✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 0c1088d. Link to the linter CI: here}

The failure is caused by a change in the output in the HDBSCAN output:

______________ [doctest] sklearn.cluster._hdbscan.hdbscan.HDBSCAN ______________
[gw0] linux -- Python 3.13.1 /usr/share/miniconda/envs/testvenv/bin/python
627 
628 Examples
629 --------
630 >>> from sklearn.cluster import HDBSCAN
631 >>> from sklearn.datasets import load_digits
632 >>> X, _ = load_digits(return_X_y=True)
633 >>> hdb = HDBSCAN(min_cluster_size=20)
634 >>> hdb.fit(X)
635 HDBSCAN(min_cluster_size=20)
636 >>> hdb.labels_
Expected:
    array([ 2,  6, -1, ..., -1, -1, -1])
Got:
    array([ 2,  6, -1, ..., -1, -1, -1], shape=(1797,))

This is caused by a change in the repr of numpy arrays after the upgrade to numpy 2.2.0:

https://numpy.org/devdocs/release/2.2.0-notes.html#changes

Personally, I like the fact that the shape is displayed by default. However, this will make the doctest difficult to work with both old and new versions numpy.

I will try to update this doctest to make it not rely on the repr of the labels_ array anymore.

I pushed 0c1088d. Please let me know what you think about it. If we merge this, we might want to cherry-pick that commit for the 1.6.X branch.

Since I pushed 0c1088d into this PR, I'll wait for others to review instead of directly merging as usual.

The work-around seems fine to me. Maybe using ... could work.

However, this will make the doctest difficult to work with both old and new versions

The approach we have taken historically is to only run the doctests in some settings, for example we could only run them in numpy >= 2.2.0.